Direct Preference Optimization (DPO) style rewards - #99

Merged

opentaco merged 12 commits into

feature/dpo-rewards

Aug 24, 2023

Commits on Jul 20, 2023

Add DirectPreferenceRewardModel
opentaco
committed
baf42de
Maint comment update
opentaco
committed
8fe9acc
Support Llama-2
opentaco
committed
01f0c3f
Update token handling
opentaco
committed
4b2d7fe
Update requirements
opentaco
committed
59b7ec0
Update requirements
opentaco
committed
d1e1f9f

Commits on Jul 21, 2023

Add trace
opentaco
committed
e8f7559
Handle nan/inf
opentaco
committed
7e8a0d0

Commits on Jul 27, 2023

Use BTLM and check max seq length
opentaco
committed
2996a22
Merge branch 'staging' into feature/dpo-rewards
opentaco
committed
4d9ef17

Commits on Aug 24, 2023

Merge branch 'staging' into feature/dpo-rewards
opentaco
authored
c1f0cc7
Add DPO to test
opentaco
committed
e1519d1